FireStats is not installed in the database
Laryngitis homøopatiske midler køb propecia edderkop bid retsmidler

PNAS Study: $90 wine tastes better than the same wine at $10

Author:movellan @ January 14th, 2008 Leave a Comment

Here is an interesting study showing that a $90 wine may taste beter than the same wine at $10

IEEE Workshop on Human Communicative Behavior. Deadline March 25th 2008

Author:movellan @ January 14th, 2008 Leave a Comment

IEEE Workshop on CVPR for Human Communicative Behavior
Deadline for paper submissions is March 25th 2008

For Bob

Author:movellan @ January 11th, 2008 Leave a Comment


Scribe Notes of TDLC Break-out 1 – Initiative 3

Author:jake @ January 5th, 2008 Leave a Comment

Scribe Notes – Breakout Session – Initiative 3

  • Mike Mozer – Theories of Attentional Control
    • Attentional Control – flexibly modulate the deployment of attention
    • Saliency maps
      • exogenous/bottom-up — doesn’t depend on particular task
      • feature-based endogenous/top-down — attention guided to task-relevant features
      • scene-based endogenous — attention guided to task-relevant features AND global-scene gist.
      • saliency P(T_x | F_x, rho) based on: environmental statistics, feature activity in neighborhood, target at location?
    • Exogenous control: break up saliency into task-independent and task-dependent terms
    • Feature-based endogenous control
    • Scene-based endogenous control: break features into global features and local features (within neighborhood of x)
    • Key to each approach is the collection of statistics of the environment
      • lifelong history
      • recent history
      • current image
      • each approach seems insufficient
    • Summary – framing attentional control from a probabilistic perspective is a big advance -
      • strength: assignment of saliency follows from probabilistic model
      • weakness: unclear of the timescale over which learning takes place
    • Questions
      • Nick Butko
      • Marni Bartlett
      • Miro Enev
      • Javier Movellan’s question – is hierarchical saliency scheme necessary?
  • Virginia de Sa – Spatiotemporal Distortions around Eye Movements
    • Premise: Amazing eye-movement finding – perrceived order of stimulus flashes is reversed if they happen before a saccade (Concetta-Morrone)
    • Why do people pereive flashes in wrong order? Replace one flash with auditory signal (not subject to these perceptual distortions)
    • Temporal compression – in a sense, swapping of ordering is compression in the extreme case
    • Flashes 0 to 50 msecs before sacccade processed more quickly than those occuring earlier
    • Saccade target blanking (move target during saccade)
    • How can this study touch on others in the center -
      • adaption to worlds (e.g., virtual environments) with different time constants
      • other eye-movement studies as they reflect on purpose of and transformations during eye movements
      • discrete sampling in other sensory-motor systems
      • Auditory/visual temporal training paradigms – how to relate timing of visual and auditory signals
    • Questions
      • Howard Poizner
      • Mike Mozer
      • Marni Bartlett – our “frame-rate” is very adaptive
      • Javier Movellan
    • Brief tour of related saccadic phenonemona:
    • Obbotson reported in SFN 2007 that responses in MT to flashes presented just prior to saccade have significantly shorter latency than those presented farther away.
    • Deubel: during saccades, perception of displacements is very poor, BUT ARE ALLEVIATED IF DISPLACED TARGET IS BRIEFLY BLANKED.
    • Virginia de Sa’s White-board notes
    • Question -
      • Michael Dabney – implications of this research for education? – implications for reading – fast readers: one saccade per word, slow readers – multiple saccades/word.
      • how much information a reader can carry between saccades related to how fast they can read
      • Howard Poizner – can we train people to speed up saccades?
      • Virginia de Sa – speed-reading – make very few saccades
    • Javier Movellan’s question
      • is the remapping possibly due to some sort of optimal way of getting a smooth view of the world?
    • Michael Dabney – effect of IQ on remapping effect?
      • all subjects in Virginia de Sa’s study were graduate students
  • Alex Simpkins (presenting Emo Todorov’s work) – Overview of Projects at Movement Control Lab
    • Active exploration
    • Reaching and decision making
    • Hierarchical optimal control
    • Modeling behavior of M1 with neural nets to control arm models for reaching tasks
    • Active exploration – project update
      • Trade-off between exploration (gather sensory info) and regular actions (exploit information)
      • Examples:
        • Eye movements
        • Exploratory finger movements
        • Whisker/ear movements
    • Problem formulation – capure this behavior within framework of stochastic optimal control & bayesian inference
      • capture both exploration and exploitation with single cost function
    • Example task – 3-d position sensor “mouse” – maps to cursor on screen
      • assume mapping is unknown, and changes over time
      • reduce task to control of diameter of the circle – damped Brownian motion alters the mapping
      • Task: make diameters of two circles equal – need to both explore transformation space and perform tracking exercise
      • very challenging task – took lots of playing around to get a reasonably but not too difficult task
    • Algorithm
      • Functional approximation scheme
      • MDP’s don’t work – state space blows up
    • Comparing performance – optimal controller to human
      • Human constantly balances exploration and exploitation
    • Future Experiments
      • Build additional robots – robot hand – compare these algorithms in action
    • Question
      • Michael Dabney – implications for teaching learners problem-solving? applications?
        • how teachers ask questions and when
        • is teacher going too slow? too fast? or should they just present more? exploration/exploitation problem!
        • can provide tool to teachers – close the loop in teaching – “know more accurately what the reality is with a student” (Alex Simpkins).
  • Javier Movellan
    • Ongoing Projects
      • Contingency Detection
      • Learning to learn
      • Hand/eye coordination during learning
    • Contingency Detection
      • How can we learn the optimal controller?
        • Use uncertainty as reinforcement signal
    • Hand-eye Coordination
      • Model end-phase saccades to fine-tune result of steering an object
    • Issues – real-time coordination of sensors and actuators
      • how humans solve tradeoff between exploration and exploitation – solves many problems simultaneously
      • there’s a danger of breaking an organism into different, separate modules
        • robotics gets around this trap since the robot as a whole has to work
      • but there’s also a robotics trap
        • “which program do you want to run”? – MIT Media Lab
    • Questions
      • Jay McClelland – Active exploration – find methods of learning that apply across domains:
      • Jay McClelland – applied exploration/exploitation to attentional context?
  • « go back